Back to timeline

Thu, October 1503:30ToolsAPI & pricingMCP & SkillsAI codingAPI & pricing guide

wshobson/agents offers a plugin marketplace across seven tools including Claude Code and Codex

Decision Brief

What changedThe wshobson/agents repository, trending on GitHub, ships a set of plugins, agents, skills, and commands usable across seven coding tools: Claude Code, OpenAI Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot, and Pi.
Why it mattersA single Markdown source generates each tool's required format via adapters, saving the cost of maintaining agents and skills per tool, but tools like Codex have an 8 KB budget and differing TOML adapter paths, so check the capability matrix before adopting.
Who should careAI coding tool users
Affected stackClaude CodeOpenAICursorCopilot
Builder actionUpdate Claude Code to the latest version
Source confidenceHigh · Official release / blog / repo

The repo contains 94 plugins (92 local, 2 external entries), with 202 agents, 184 skills, and 105 commands in the local source. Claude Code installs directly via the plugin marketplace; Cursor uses the committed marketplace registry; Codex CLI supports both a native marketplace and generated TOML agent/command-skill adapter paths; OpenCode, Antigravity CLI, GitHub Copilot, and Pi require cloning the repo and running make install-*, and Pi agents also depend on its sub-agent extension. Skills can be installed individually with gh skill or npx skills, requiring GitHub CLI 2.90+ or Node.js and npm. Claude Code developers can register the marketplace with /plugin marketplace add and install plugins, reusing the same Python, JavaScript, code review, and security scanning components; teams maintaining multiple toolchains must run the matching install command per tool and accept that model aliases (opus, sonnet, haiku, fable, inherit) are mapped by adapters to each tool's model settings, with some inherit entries pinned to defaults. The repo provides make generate-all, make validate STRICT=1, make garden, and plugin-eval static checks; the LLM-judge and Monte Carlo layers are experimental, not aligned with human labels, and consume quota on the account tied to ANTHROPIC_API_KEY, otherwise using Claude Code login.

Summary basis: full article readCompiled from the source scope noted above; the original remains authoritative.

Sources

  • Skill Radar(GitHub 趋势)

    Trending hands-on MCP servers, agent skills, and AI-coding tools discovered from GitHub search momentum.

  • GitHub:wshobson/agents

Related intel

留言

登入后即可留言,和其他 builder 交换实测心得。

还没有留言,抢头香。